I'm referring to Otsu's thresholding, and saw this from his paper:
enter image description here

When he says he's trying to maximize the between class variance of thresholds k* 1 and 2, is it the threshold with maximum value that is chosen? Is that what the "max" means? If it was just about selecting one threshold, why is it called multi-level thresholding?

On the other hand, if multiple thresholds are chosen, then why do we need to find the "max" of it?

ps: I'm trying to implement Otsu's method in a computational intelligence algorithm (Differential Evolution and Particle Swarm Optimization), and not a single research paper has explained clearly, how the thresholds are to be used for multi-thresholding. Not even the original Otsu paper. Also, after implementing the algorithm, the image with the best fitness threshold is a different (much darker) threshold than what Matlab's greythresh threshold returns.

  • $\begingroup$ Please provide a complete citation for Otsu. Many of us are generalists $\endgroup$
    – user28715
    Aug 30, 2017 at 18:09
  • $\begingroup$ Oh ok...I've added links to the original Otsu paper. I thought this was the right forum to ask this question. Didn't find any separate image processing StackExchange forum. $\endgroup$
    – Julia
    Aug 31, 2017 at 16:30
  • $\begingroup$ @Anon, Many are looking for a separate Image Processing Community (Or name this one Signal and Image Processing). $\endgroup$
    – Royi
    Aug 31, 2017 at 17:09

1 Answer 1


Using k1 and k2 you can calculate a between class variance value using the Otsu formula. However, there may be many combinations of k1 and k2 for which you have to calculate between class variance values to see which of them gives the highest between class variance value. That's what 'max' means. So whichever k1 and k2 gives you the highest between class variance, you choose those values of k1 and k2 as your thresholds. Same can be done if you want more thresholds.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.